2 Inheritance

Objective

Building inheritance enables a single source of truth that reduces review burden.

3 Overview

To ensure consistency in clinical study reporting deliverables, a biometrics organization has established standards, business rules, and templates.

Building AI-prompt-ready templates that can interact with GenAI in a predictable way is critical. Template inheritance allows organizations to create reusable keyword definitions that can be shared across therapeutic areas and studies, while maintaining study-specific flexibility.

3.1 Inheritance Hierarchy

To enable GenAI to produce accurate and consistent deliverables, we must maximize the use of organization and therapeutic area (TA) level templates, ensuring that study analysis and reporting plans inherit this knowledge.

Organization Templates
    ↓
TA Templates
    ↓
Study Plans

4 Example

For clinical analysis and reporting specifically, the template can be organized as a collection of keywords in different categories including population, observation, and parameter, etc.

4.1 Organization or TA Template

For simplicity, we do not differentiate between organization and TA templates. A template can be as simple as a keyword name and its label that is expected to be used across TLFs and documentation (e.g., apat and itt).

organization.yaml

population:
  - name: apat
    label: "All Participants as Treated"
  - name: itt
    label: "Intention-to-Treat"

parameter:
  - name: any
    label: "Any Adverse Event"
  - name: rel
    label: "Related Adverse Events"
  - name: ser
    label: "Serious Adverse Events"

4.2 Study Plan Inheritance

Reference the template “organization.yaml”, so the knowledge will be inherited while building the study analysis plan.

plan_xyz123.yaml

study:
  name: "XYZ123"
  template:
    - organization.yaml

To make a keyword useful, we can define a filter following an enhanced SQL syntax. For example, adae:trtemfl == "Y" reads as the ADAE dataset trtemfl flag equal to Y.

plan_xyz123.yaml

# Study-specific parameter filters
parameter:
  - name: any
    filter: "adae:trtemfl == 'Y'"
  - name: rel
    filter: "adae:trtemfl == 'Y' and adae:aerel in ['RELATED', 'PROBABLY RELATED']"

4.3 Field-Level Inheritance

The system performs intelligent field-level merging:

4.3.1 Example Resolution

# Template: organization.yaml
parameter:
  - name: any
    label: "Any Adverse Event"

# Study: plan_xyz123.yaml
parameter:
  - name: any
    filter: "adae:trtemfl == 'Y'"

# Final merged result:
parameter:
  - name: any
    label: "Any Adverse Event"        # From template
    filter: "adae:trtemfl == 'Y'"     # From study

4.4 Multiple Templates

Templates are loaded in order with later templates able to override earlier ones.

study:
  template:
    - organization.yaml
    - ta_oncology.yaml
    - safety_common.yaml

4.4.1 Conditional Inheritance

Study plans can completely override template definitions when needed:

# Study completely replaces template definition
parameter:
  - name: ser
    label: "Study-Specific Serious Adverse Events"
    filter: "adae:aeser == 'Y'"

4.5 Benefits

4.5.1 1. Consistency

Organization-wide keyword standards
Consistent labeling across studies
Regulatory compliance templates

4.5.2 2. Efficiency

Reduce duplication
Faster study setup
Standardized filters

4.5.3 3. Maintenance

Update templates to affect all studies
Version control for standards
Clear inheritance chain

4.5.4 4. Flexibility

Study-specific customizations
Override capability when needed
Gradual migration of standards

This inheritance system provides the balance of standardization and flexibility for large-scale organizations.

--- title: "Inheritance" --- ::: callout-tip ## Objective Building inheritance enables a single source of truth that reduces review burden. ::: # Overview To ensure consistency in clinical study reporting deliverables, a biometrics organization has established standards, business rules, and templates. Building AI-prompt-ready templates that can interact with GenAI in a predictable way is critical. Template inheritance allows organizations to create reusable keyword definitions that can be shared across therapeutic areas and studies, while maintaining study-specific flexibility. ## Inheritance Hierarchy To enable GenAI to produce accurate and consistent deliverables, we must maximize the use of organization and therapeutic area (TA) level templates, ensuring that study analysis and reporting plans inherit this knowledge. ``` Organization Templates ↓ TA Templates ↓ Study Plans ``` # Example For clinical analysis and reporting specifically, the template can be organized as a collection of keywords in different categories including population, observation, and parameter, etc. ## Organization or TA Template For simplicity, we do not differentiate between organization and TA templates. A template can be as simple as a keyword name and its label that is expected to be used across TLFs and documentation (e.g., `apat` and `itt`). **`organization.yaml`** ```yaml population: - name: apat label: "All Participants as Treated" - name: itt label: "Intention-to-Treat" parameter: - name: any label: "Any Adverse Event" - name: rel label: "Related Adverse Events" - name: ser label: "Serious Adverse Events" ``` ## Study Plan Inheritance Reference the template "organization.yaml", so the knowledge will be inherited while building the study analysis plan. **`plan_xyz123.yaml`** ```yaml study: name: "XYZ123" template: - organization.yaml ``` To make a keyword useful, we can define a `filter` following an enhanced SQL syntax. For example, `adae:trtemfl == "Y"` reads as the `ADAE` dataset `trtemfl` flag equal to `Y`. **`plan_xyz123.yaml`** ```yaml # Study-specific parameter filters parameter: - name: any filter: "adae:trtemfl == 'Y'" - name: rel filter: "adae:trtemfl == 'Y' and adae:aerel in ['RELATED', 'PROBABLY RELATED']" ``` ## Field-Level Inheritance The system performs intelligent field-level merging: ### Example Resolution ```python # Template: organization.yaml parameter: - name: any label: "Any Adverse Event" # Study: plan_xyz123.yaml parameter: - name: any filter: "adae:trtemfl == 'Y'" # Final merged result: parameter: - name: any label: "Any Adverse Event" # From template filter: "adae:trtemfl == 'Y'" # From study ``` ## Multiple Templates Templates are loaded in order with later templates able to override earlier ones. ```yaml study: template: - organization.yaml - ta_oncology.yaml - safety_common.yaml ``` ### Conditional Inheritance Study plans can completely override template definitions when needed: ```yaml # Study completely replaces template definition parameter: - name: ser label: "Study-Specific Serious Adverse Events" filter: "adae:aeser == 'Y'" ``` ## Benefits ### 1. Consistency - Organization-wide keyword standards - Consistent labeling across studies - Regulatory compliance templates ### 2. Efficiency - Reduce duplication - Faster study setup - Standardized filters ### 3. Maintenance - Update templates to affect all studies - Version control for standards - Clear inheritance chain ### 4. Flexibility - Study-specific customizations - Override capability when needed - Gradual migration of standards This inheritance system provides the balance of standardization and flexibility for large-scale organizations.